That makes sense considering there are those checks for ignoring 1 missing version.
On Fri, Apr 29, 2016 at 6:53 AM, Steve Rowe <sar...@gmail.com> wrote: > Anshum, > > TL;DR: When there is only one release in flight, I think it’s okay to run > addVersion.py on all branches at the start of the release process for all > types of releases. > > When we chatted last night I said backcompat index testing was a problem > on non-release branches in the interval between adding a not-yet-released > version to o.a.l.util.Version and when a backcompat index is committed on > the branch. I was wrong. > > Here are the places where there are back-compat coverage tests: > > 1. smokeTestRelease.py's confirmAllReleasesAreTestedForBackCompat() will > succeed until release artifacts have been published - see > getAllLuceneReleases() for where they are scraped off the lucene release > list page on archive.apache.org. So back-compat indexes should be > generated and committed as soon as possible after publishing artifacts. > > 2. backward-codec’s TestBackwardsCompatibility.testAllVersionsTested() > will still succeed if a single version is not tested. Here’s the code: > > // we could be missing up to 1 file, which may be due to a release that > is in progress > if (missingFiles.size() <= 1 && extraFiles.isEmpty()) { > > The above test could be improved by checking for the presence of published > release artifacts for each release like smokeTestRelease.py does, and then > not requiring the backcompat index be present for those that are not yet > published; this would allow for multiple in-flight releases. > > Steve > > > On Apr 28, 2016, at 10:44 PM, Anshum Gupta <ans...@anshumgupta.net> > wrote: > > > > I've updated the "Update Version Numbers in the Source Code" section on > the ReleaseToDo page. It'd be good to have some one else also take a look > at it. > > > > Here is what I've changed (only bug fix release): > > * Only bump up the version on the release branch using addVersion.py > > * Don't bump it up on the non-release versions in case of bug fix > release. > > * As part of the post-release process, use the commit hash from the > release branch version bump up, to increment the version on the non-release > branches. > > > > I thought we could do this for non bug-fix releases too, but I was > wrong. Minor versions need to be bumped up on stable branches (and trunk) > because during the release process for say version 6.1, there might be > commits for 6.2 and we'd need stable branches and master, both to support > those commits. > > We could debate about not needing something like this for major versions > but then I don't think it's worth the pain of different release processes > for each branch but I'm not stuck up with this. > > > > > > On Thu, Apr 28, 2016 at 5:31 PM, Anshum Gupta <ans...@anshumgupta.net> > wrote: > > That's fixed (about to commit the fix from LUCENE-7265) thought. > > > > While discussing the release process, Steve mentioned that we should > document the failing back-compat index test on the non-release branches due > to the missing index for the unreleased version. > > On discussing further, he suggested that we instead move the process of > adding the version to non-release branches as a post-release task. This > way, we wouldn't have failing tests until the release goes through and the > back-compat indexes are checked in. > > > > We still would have failing tests for the release branch but there's no > way around that. > > > > So, I'll change the documentation to move those steps as post-release > tasks. > > > > > > On Thu, Apr 28, 2016 at 11:40 AM, Anshum Gupta <ans...@anshumgupta.net> > wrote: > > Seems like LUCENE-6938 removed the merge logic that used the change id. > Now the merge doesn't happen, and there's no logic that replaces it. > > > > I certainly can do with some help on this one. > > > > On Thu, Apr 28, 2016 at 11:24 AM, Anshum Gupta <ans...@anshumgupta.net> > wrote: > > Just wanted to make sure I wasn't missing something here again. While > trying to update the version on 5x, after having done that on 5.5, using > the addVersion.py script and following the instructions, the command > consistently fails. Here's what I've been trying to do: > > > > python3 -u dev-tools/scripts/addVersion.py --changeid 49ba147 5.5.1 > > > > Seems like addVersion.py is broken for minor version releases so I'd > need some help with someone who has a better understanding of python than I > do. I observed that 5.5.1 Version gets added to Version.java but also gets > marked as deprecated. > > > > > > > > On Thu, Apr 28, 2016 at 9:27 AM, Anshum Gupta <ans...@anshumgupta.net> > wrote: > > Too much going on! Thanks Yonik. > > I'll start working on the RC now. > > > > NOTE: Please don't back port any more issues right now. In case of > exceptions, please raise them here. > > > > On Thu, Apr 28, 2016 at 9:09 AM, Yonik Seeley <ysee...@gmail.com> wrote: > > On Thu, Apr 28, 2016 at 12:04 PM, Anshum Gupta <ans...@anshumgupta.net> > wrote: > > > Thanks. I'm waiting for the last back port of SOLR-8865. > > > > It should be already be there... I closed it yesterday. > > -Yonik > > > > --------------------------------------------------------------------- > > To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org > > For additional commands, e-mail: dev-h...@lucene.apache.org > > > > > > > > > > -- > > Anshum Gupta > > > > > > > > -- > > Anshum Gupta > > > > > > > > -- > > Anshum Gupta > > > > > > > > -- > > Anshum Gupta > > > > > > > > -- > > Anshum Gupta > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org > For additional commands, e-mail: dev-h...@lucene.apache.org > > -- Anshum Gupta